In the quiet hours of a July night, three researchers from a small Indian firm sat down with an AI model built by one company and used it to breach the defenses of another — gaining access to private conversations, employee accounts, and connected workspaces in under three days. The incident is less a story about one hack than about a threshold being crossed: the tools that once made sophisticated cyberattacks the province of nation-states are now available to small teams working overnight. What took months now takes days, and the industry racing to build these tools has not yet reckoned with
Indian hackers breach OpenAI using rival Anthropic model, exposing user data risks
OpenAI users' private conversations and connected accounts (email, Slack) were exposed to unauthorized access, including data from OpenAI employees.